Runs posterior- and prior–predictive analyses and returns summaries of performance
Source:R/posterior_predictive_analysis.R
posterior_predictive_analysis.RdThis function takes in a model object, the observed data, and the configuration file used to produce the real-data stan data object.
If running a posterior-predictive analysis, it then fits the model, extracts generated quantities from some number of those fits, and runs new analyses on the simulated data. Lastly it returns summaries of estimation performance.
If running prior-predictive analyses, it runs under the prior to get the joint prior distribution. Then it extracts generated quantities from some number of those fits, and runs new analyses on the simulated data. Lastly it returns summaries of estimation performance.
Usage
posterior_predictive_analysis(
model,
real_data,
config,
draw = NULL,
real_data_fit = NA,
ci_width = 0.89,
mclapply_cores = NA,
under_prior = FALSE,
...
)Arguments
- model
a CmdStanModel object defining the model
- real_data
observed data, formated for stan
- config
the output of get_config_vals()
- draw
vector, which draws from the MCMC object do we want to run PPS analyses for? If NULL, runs one analysis per sample. This can take a long time and a lot of space!
- real_data_fit
CmdStanMCMC object, optional argument providing a previous real-data model fit
- mclapply_cores
integer specifying number of threads to use for posterior-predictive-analysis-level parallelization, separate from stan parallelization
- under_prior
boolean, if TRUE prior-predictive analysis is run, not posterior
- ...
further arguments passed to model$sample(), e.g. # chains
- pps_mcmc_fits
list of CmdStanMCMC objects, optional argument providing previously-run pps analyses (for re-summarizing)